Abstract
Many efforts are made to promote soil conservation in developing countries. This paper compares the effect of two programs promoting soil conservation in Peru on the adoption decision of households. One program applies a top-down approach with soil conservation as its core activity. The other program applies a participatory approach, offering a portfolio of activities in order to improve rural livelihoods. The decisions on participation and adoption are estimated with a trivariate probit model. The results show that each program attracts different types of households and achieves different outcomes in terms of soil conservation. (JEL C35, O13)
I. Introduction
In many developing countries there is widespread concern about the threat of soil erosion, potentially resulting in decreased agricultural production and consequently affecting the well-being of farm households (Barbier 1995; Scherr 1999). Therefore, governments and development programs have been undertaking various initiatives to prevent soil erosion and enhance sustainable agriculture by promoting soil conservation practices. However, despite all these efforts, adoption rates of soil conservation practices are often disappointing (Lutz, Pagiola, and Reiche 1994; Rist and Martin 1991; Sanders and Cahill 1999).
The implementation of soil conservation practices, in particular structural practices such as terraces, potentially requires considerable investments in capital and labor, while immediate benefits (i.e., positive impacts on production or profit) for farm households tend to be limited or uncertain. The adoption decision can thus initially be considered as a “risky choice problem” (Marsh 1998; Marra, Pannell, and Ghadim 2003). Programs promoting soil conservation often use incentives, in addition to extension, to overcome adoption constraints and to reduce the risk of adopting a new technology (Nowak 1987). An incentive can be defined as “any inducement on the part of an external agency, meant to both allow and motivate the local population, be it collectively or on an individual basis, to adopt new techniques and methods aimed at improved natural resource management” (Zaal, Laman, and Sourang 1998). Common incentives, particularly those used in conventional top-down soil conservation programs, are direct support such as food-for-work, tools-for-work, or free distribution of seeds and fertilizers. These conventional soil conservation programs have often been criticized for not obtaining sustainable results in the long term. Many examples are known where incentives induced changes in land management practices over the short term, but farm households returned to their old habits, abandoning the new practices, as soon as these incentives were withdrawn. Farm households were only interested in the incentive itself (e.g., food, cash, agricultural equipment), and not in the accompanying soil conservation technology (Giger 1999; Hellin and Haigh 2002; Pretty and Shah 1997; Sanders and Cahill 1999). Despite these adverse effects, some programs continue using direct incentives to achieve quick results, without paying much (time-consuming) attention to the sustainable effect of the program (Bunch 1999). However, it has been increasingly suggested that participatory approaches to soil conservation are likely to obtain more sustainable results in the long term. Common elements of successful soil conservation programs include the following (Bunch 1995; Pretty and Shah 1997):
Recognition of the value of local knowledge and technologies
Farmers at the centre of innovation and adaptation process of new technologies, enabling them to experiment on a small scale, reducing the risk of adoption
Use of technologies that rely on locally available resources and deliver shortterm benefits
Training of farmers as experts/extension agents
Encouragement of collective action through formation of farmer groups
Creation of a supporting/enabling institutional environment focusing on local needs and capabilities
The aim of this paper is to better understand the impact of soil conservation promotion campaigns on adoption behavior of farm households. The Andes is chosen as the research area because of its problems with soil erosion and the extensive experience with soil conservation interventions. Two governmental programs promoting soil conservation in Peru are discussed in this paper. The national program Proyecto Nacional de Manejo de Cuencas Hidrograficas y de Conservacion de Suelos (PRONAMACHCS), applying a top-down approach, and the participatory pilot program Manejo de Recursos Naturales en la Sierra Sur (MARENASS). The paper assesses whether these two different soil conservation promotion campaigns have a causal effect on soil conservation adoption by farm households.
II. Soil Conservation Programs in Peru
Though many nongovernmental organizations (NGOs) promote soil conservation throughout the Andes, only two governmental programs are considered in this paper, PRONAMACHCS and MARENASS, as these two programs were promoting soil conservation in the research area. Table 1 summarizes the main distinguishing features of the two programs.
Main Distinguishing Features of Programs Pronamachcs and Marenass
The Peruvian government started a still ongoing, large-scale soil conservation program in 1981PRONAMACHCS. Initially, PRONAMACHCS promoted mainly soil conservation, but other activities were incorporated over time, such as reforestation, road construction, and support for agricultural production (Vogel, Rojas Pineda, and Sallo Pumacahua 2004). According to PRONAMACHCS, lack of knowledge is the principal restriction for farmers to implement soil conservation practices. Knowledge transfer (see Rogers 1995) is therefore seen as the solution to promote soil conservation: by involving farm households in soil conservation activities, they learn how to implement these practices and at the same time they can observe the outcomes (Chang-Navarro 1986). Technical staff of PRONAMACHCS organizes activities, such as construction of terraces or reforestation, once a week in each community. This is normally done in collaboration with a “promoter,” a local farmer appointed by PRONAMACHCS who acts as an intermediary between the community and PRONAMACHCS. In general, a technical engineer is responsible for the implementation of soil conservation practices. PRONAMACHCS' strategy is to start the implementation of soil conservation practices in the upper parts of the catchment and then work their way downward. The argument behind this strategy is that there is little use in implementing soil conservation practices downhill if surface runoff, causing soil erosion, coming from uphill areas has not been reduced.
The motivation of the program's technical staff largely determines the success of the program. Heredia (1997) found that staff members were concerned about achieving their (quantitative) targets in order to ensure their job, rather than to promote sustainable solutions for natural resource management. The top-down structure of PRONAMACHCS, hindering the adaptation of interventions to local circumstances, and the emphasis on quantitative targets created a structure that encouraged short-term outputs rather than sustainable management of natural resources in the long run (Heredia 1997). Therefore, PRONAMACHCS was reorganized in 2001, and it now increasingly promotes active participation of farm households in decision making and planning of program activities, as well as more integrated approaches toward natural resource management (Vogel, Rojas Pineda, and Sallo Pumacahua 2004).
In order to incentivize participants, PRONAMACHCS provided food-for-work in the past, which was targeted at poor farm households with low food security. Nowadays PRONAMACHCS provides tools to farm households as an incentive for the implementation of soil conservation practices. These tools, such as shovels, pickaxes, and wheelbarrows, are typically required for the construction of terraces and infiltration ditches. Each farm household that “supplies'' at least one household member each week to participate in the activities receives tools at the end of the year. In the year 2000, PRONAMACHCS worked in 866 catchments in the Peruvian Andes, and it was estimated that in this year 232,772 households were reached. The area with soil conservation practices was estimated at 38,920 ha, with half of this area constituting slow-forming terraces, a third of it infiltration ditches, and the remaining area new or rehabilitated bench terraces (PRONAMACHCS 2000).
In 1998, the Peruvian government launched a pilot program, MARENASS, which terminated at the end of 2003. MARENASS worked in 360 communities in total, reaching about 33,000 households (Zutter 2004). MARENASS worked in a community for a maximum of four years. The main goal of MARENASS was to enhance the capacity of peasant communities to improve the productivity of natural resources to overcome poverty. At the start ofthe intervention, MARENASS presented itself during village assemblies, and if the community agreed on joining the program, a contract was signed. The participating communities had to open a bank account, to which funds were transferred to finance program activities. As such, MARENASS transferred decision making and responsibility for planning and financial resources to the communities. Participation of community members was on a voluntarily basis. MARENASS also used local “promoters” (promotor communal) as an intermediary between the program and the community. These promoters were appointed by the village assembly and were typically local farmers with leadership and management skills. The promoter played a central role in motivating participants and achieving successful outcomes.
MARENASS promoted a range of activities such as improvement of grassland, soil conservation, house improvement, horticulture, construction of sanitary facilities, animal breeding, handicraft, public works at the community level, empowerment, and improving community dynamics. MARENASS participants decided themselves in which program activities they wanted to participate. The program staff acted as facilitators rather than extension officers, helping participants to organize themselves, to get access to knowledge and capital, and to set up trainings. MARENASS encouraged farmer-to-farmer extension, where innovative and experienced farmers were hired by the communities to explain new or improved technologies to their peers. Farm households also decided themselves on the implementation of soil conservation practices, including the type of practice and location.
The main distinctive features of MARENASS were the participatory, demand-driven approach, the emphasis on empowerment, and the organization of farmer competitions. In order to incentivize participants, farmer competitions were organized at the community level (Cabero and Van Immerzeel 2007), with farm households competing against each other in their uptake performance of new technologies. The uptake was typically measured in terms of quantity rather than quality; for example, in the case of soil conservation the uptake was evaluated by the area with newly constructed terraces, but the quality or location (“fit for purpose”) of the terraces was not taken into account. Farm households with the best uptake received cash awards. In the research area, farm households decided to enter the farmer competitions as groups, since they deemed it unfair if some participants would not receive any reward despite putting effort into the competitions. By competing between a limited number of groups, every participant was assured of some reward, as all groups received cash awards according to their collective performance in the competitions. The money was either reinvested in group activities or divided among the participants.
The main soil conservation practices promoted by the two programs were slow-forming terraces, bench terraces, and infiltration ditches. Slow-forming terraces consist of bunds along the contour that are stabilized with vegetation. Ongoing infield soil erosion and sedimentation processes modify the slope over time. Bench terraces consist of a series of alternating platforms and risers. Bench terraces modify the slope to enable maximum infiltration of rainwater, and as a consequence they reduce runoff and erosion. Infiltration ditches are ditches along the contour that intercept runoff in order to let it infiltrate into the soil. Bench terraces require the most labor input for construction, infiltration ditches the least.
An immediate benefit of the practices is the improved water retention, reducing the risk of crop failure due to drought. Furthermore, surface irrigation can be applied on bench terraces during the dry season, enabling a second short growing season. Benefits over the long term are the reduction of soil erosion and, subsequently, the maintenance of soil fertility. A cost-benefit analysis of bench terraces considering a 10-year period revealed that these practices are financially attractive to farm households if use is made of the improved cropping conditions by cultivating crops with a higher value. Opportunity costs for labor appeared the most determining factor of profitability, whereas the incentives provided by the programs had little effect (Posthumus and De Graaff 2005).
III. Conceptual Framework for Soil Conservation Adoption
Three main paradigms can be distinguished in the theory of adoption of soil conservation practices: the economic constraint paradigm, the innovation-diffusionadoption paradigm, and the adopter perception paradigm (Adesina and Zinnah 1993). Table 2 summarizes the determining factors for adoption according to these paradigms.
Mainstream Theoretical Models on Adoption of Soil Conservation Practices
The economic constraints paradigm assumes that resource endowments are asymmetrically distributed across farm households, determining the observed pattern of adoption (Adesina and Zinnah 1993; Negatu and Parikh 1999). Potential economic constraints are (Foltz 2003) natural resource endowments (e.g., land), capital scarcity, learning costs associated with implementing a new technology, and risk attitude. The economic constraints paradigm assumes that farm households strive for profit or utility maximization. The strength of this paradigm lies in the role of profitability that motivates innovation or adoption. However, farm households in developing countries often opt for acceptable rather than maximum profits, as nonfinancial variables (e.g., leisure, traditions, environmental protection) are also important in their decision making (Ellis 1993).
The innovation-diffusion-adoption paradigm is based on the innovation-diffusion theory of Rogers (1995). According to this paradigm, access to information is the key factor determining adoption decisions. Assuming the innovation is appropriate, the problem of technology adoption is reduced to communicating information to potential adopters (Adesina and Zinnah 1993). The strength of the innovation-diffusion-adoption paradigm is the recognition that adoption is a multistage process of collecting information, revising opinions, and reassessing decisions (Feder, Just, and Zilberman 1982; Marsh 1998). However, this paradigm does not take individual characteristics of the adopter into account.
In the case of soil conservation, technologies are primarily designed to benefit the environment rather than increase profit, although there are technologies than can achieve both. But attitude toward environmental stewardship, in addition to financial considerations, plays an important role in the decision making regarding adoption of environmental technologies. This attitude is determined by human values and beliefs (Lynne, Shonkwiler, and Rola 1988). Ervin and Ervin (1982) incorporated this concept into the adopter perception paradigm arguing that the adoption process of soil conservation starts with the perception (or recognition) of the potential adopter that there is a soil erosion problem. This perception is determined by personal factors (e.g., human values, education, and experience) as well as physical factors of the land and institutional factors (e.g., awareness raising through extension).
Based on any of these three paradigms or a mixture them, empirical studies have found a number of significant factors determining adoption of soil conservation that can be categorized into physical, personal, economic, and institutional factors (see Table 3). Many of these are identical to the factors identified by the theoretical models.
Empirical Findings on Decisive Factors in Adoption of Soil Conservation Practices
The current theoretical and empirical literature recognizes that adoption behavior is complex and requires a blend of theoretical models (Upadhyay et al. 2003). Virtually all adoption decisions are preceded by an information acquisition period, also called an awareness or learning period (Saha, Love, and Schwart 1994; Adegbola and Gardebroek 2007). Generation and distribution of knowledge—for example, through extension and training—is an important factor in this process and is often induced by governmental programs, NGOs, or extension services.
Soil conservation programs thus have an important impact on the adoption process of soil conservation practices, as these programs typically provide or facilitate the information on the new technology, either using a knowledge transfer approach (in accordance with the innovation-diffusionadoption paradigm) or a participatory technology development approach (adopter perception paradigm), and often try to reduce the investment costs or adoption risk by providing incentives (economic constraints paradigm).
The theoretical background of this study is based on a combination of the three paradigms discussed earlier in this section. Important variables hypothesized to affect adoption of soil conservation practices in the research area are based on the studies mentioned in Table 3. In the empirical model it is recognized that program participation decisions may correlate with unobserved factors that also determine adoption of soil conservation practices. Neglecting this in estimation would yield biased estimates. Since our adoption equation is a nonlinear binary choice model, the often-applied Heckman two-step correction is not correct, and it is preferred to estimate the equations for the different decisions jointly using multivariate probit models (Greene 1998). Existing studies usually consider one extension program aiming at adoption of a particular technology (e.g., Amsalu and De Graaff 2007), leading to a bivariate probit model, which is straightforward to estimate (Greene 2003, 710-14). The unique character of this study is that it analyzes the impact of two rather different extension programs on adoption of soil conservation practices. Therefore, a trivariate probit model (Greene 2003, 714-15) is used to model participation in PRONAMACHCS and MARENASS and adoption of soil conservation practice jointly. Since standard maximum likelihood methods are not appropriate to estimate this model the Geweke-Hajivassiliou-Keane (GHK) simulator is used (Greene 2003, 714, 932-33).
IV. A Model to Explain Program Participation and Soil Conservation Adoption
At the initial stages of the adoption process, farm households have to decide on three issues: participation in the PRO-NAMACHCS program, participation in the MARENASS program, and adoption of soil conservation practices. Program participation means that a farm household has contact at least once a year with field staff of the program concerned. In principle, all farm households have equal access to PRONAMACHCS and MARENASS, and the decision to participate is made at the household level. The two programs work independently of each other and do not encourage participants to participate in the other program as well. Once a household participates in one or both of these programs, its decision to adopt soil conservation practices might be influenced by the incentives provided by the program(s). Both programs disseminate information about soil conservation practices and provide various incentives (tools or cash) in order to stimulate farm households to adopt these. Adoption is defined as the implementation of soil conservation practices by the farm household on at least one of its fields. Note that program participation is not a prerequisite for adoption, as nonparticipants can also decide to adopt soil conservation practices. Nevertheless it is expected that program participation enhances the adoption decision.
Modeling of such decision problems is often cast in a limited dependent variable framework. If there is information available on the quantity of practices installed (e.g., number ofhectares), then a Tobit or sample selection model is the preferred choice, since such models can explain both the decision to adopt and the extent of adoption, allowing for different variables and different marginal effects in both processes (Greene 2003, 770, 780-87). Examples dealing with adoption of soil conservation practices are provided by Baidu-Forson (1999) and Lynne, Shonkwiler, and Rola (1988). In our study we do not have accurate information on the extent of adoption under the different programs, so we can only analyze adoption decisions and not the extent of adoption. A large proportion of the households implemented soil conservation practices under direction of both programs, and it was impossible to separate the amount of effort put into the implementation of soil conservation practices for each program for these cases. Therefore, a binary-choice model is more appropriate for this case study.
Binary-choice models assume that individuals are faced with a choice between two alternatives (Greene 2003, 668-70). The underlying assumption is that the adoption behavior of an individual is derived from maximization of expected utility subject to constraints such as human, financial, and natural capital (Feder, Just, and Zilberman 1982). It is assumed that the individual compares the new technology or strategy with the current one, and if the individual expects that the new technology or strategy increases its utility level, he will adopt the new one (Batz, Janssen, and Peters 2003). However, using a binary-choice model has some drawbacks, as information on adoption behavior might get lost, such as effort or intensity of use and the changing adoption process over time. A continuous dependent variable would be more appropriate to estimate the effort or intensity of use, whereas panel data would be needed to model the dynamics of the adoption process through time.
The purpose of this paper is to analyze the impact of the decision regarding program participation on the decision regarding adoption. Because of potential relationships between the three decision problems, a trivariate probit model is used. A trivariate probit model enables simultaneous estimation of the decisions on participation in PRONAMACHCS, participation in MARENASS, and adoption of soil conservation practices. More importantly, it can be shown that parameter estimates in such a model are not affected by self-selection bias (Greene 2003, 714-15), which may arise if unobserved determinants of program participation also affect adoption.
For each decision problem j, households compare the expected utility of a positive decision (“yes”) with the expected utility of a negative decision (“no”). It is assumed that the two programs influence the expected utility of soil conservation either by changing awareness or attitude, or by providing incentives that are attractive to farm households. Note that expected utility can also refer to an expected sum of discounted future utilities. The difference between these unobserved utilities is denoted by the latent variable 
. Although the variables 
are not observable, they are assumed to depend linearly on a set of explanatory variables Xj and an unobserved residual term εj, namely, 
, where βj is a vector of parameters. Moreover, the actual decisions on program participation and adoption of soil conservation practices are observed, so we can define variables yj that equal one for positive decisions and zero for negative ones. This leads to the following trivariate probit specification:
[1]where subscript P refers to participation in PRONAMACHCS, subscript M to participation in MARENASS, and subscript A to adoption of soil conservation practices. Wilde (2000) showed that such a system of three equations defines eight unique probabilities (one is determined by the adding-up restriction; see the Appendix), exactly enabling estimation of three intercepts, two program participation parameters, and three correlation parameters ρ, so that even with no exogenous regressors the system is exactly identified. Addition of exogenous explanatory leads to overidentification of the model and thus improves its estimation. The covariance terms ρ indicate that equations for program participation and adoption may be related via their residual terms; that is, they may have measurement errors, shocks, or missing covariates in common. The equation for adoption indicates that besides explanatory variables XA, also participation in one or both programs, represented by dummy variables yP and yM, is expected to affect adoption as argued above. However, these variables are potentially correlated with residual term εA, which in standard regression models leads to biased parameter estimates. The reason for this bias is that farm households can self-select into the group of program participants (or not), causing a self-selection bias (Edmonds 1999). Unobserved variables influencing program participation might influence the adoption decision as well. In other words, residuals εP and εM might be correlated with residuals εA. However, since yP and yM correlate with εP and εM, both program participation dummies also correlate with εA. The implication of this self-selection bias might be that differences between program participants and nonparticipants are not due to the program impacts, but due to this (self-) selection process.
Greene (1998, 2003, 715-16) has shown that in estimating a bivariate probit model with endogenous binary regressors using full information maximum likelihood, one can ignore the endogenous nature of the binary regressors and proceed as if there were no endogeneity problem. The reason for this is that the estimation procedure is based on maximizing a log likelihood, which is based on the joint probability distribution defined by the different combinations of the binary variables, whereas least squares or generalized method of moments estimation is based on sample moments that do not necessarily converge to zeros.1 Although in our model there are two potentially endogenous regressors (yP and yM) in the adoption equation, Greene's argument can easily be extended to the trivariate probit model with binary regressors as used in this study. See the Appendix for the definition of the likelihood function. The trivariate probit model was estimated using the triprobit procedure that was written by Terracol (2002), which is based on the GHK simulator.
V. Methodology
Research Area
The data was collected in the district of Pacucha (department of Apurimac) in the southern part of the Peruvian Andes. The district Pacucha covers 170 km2 and is situated near the rural town Andahuaylas, connected by an unpaved road. The altitude ranges from 3,000 to 4,000 m above sea level. Climate is temperate, with an average annual rainfall of 700 mm. The main crops cultivated are maize and potato, and to a lesser extent cereals. Soils are mainly loam or silt loam and susceptible to erosion.
Despite widespread erosion, soil conservation practices are not widely implemented. In contrast to other parts of the Andes, there are no ancient terraces famously built by the Incas, as this region was inhabited by the Chankas, enemies of the Incas. During the 1980s and early 1990s, this area was controlled by guerrillas (Shining Path). Because of its remoteness and political instability NGOs and governmental programs started interventions in this area only from the mid-1990s onward. Lack of technical capacity and knowledge is thus a major factor explaining the limited implementation of soil conservation practices, as well as limited access to credit and markets. Indeed, local farmers commented that lack of knowledge and limited program assistance were the main causes of the limited use of soil conservation practices (Posthumus 2005). PRONAMACHCS started its activities in this area in 1995, MARENASS in 1998.
Farm households undertake various income-generating activities such as agriculture, crafts, running small businesses, or working as laborers. The agricultural production is mainly for own consumption, but surpluses are sold on the local market. Most farm households dedicate part of their time to nonfarm activities to earn some extra cash income. The 2007 national census registered 3,284 households in Pacucha, of whom 88% lived in the rural area (INEI 2007).
In this region, the land belongs to the peasant communities, and each farm household has permanent private user rights of its fields in proximity to its home. These fields are intensively cultivated by the farm household, in particular if there is access to irrigation water. Fields situated in the upper parts of the catchment, and thus at a larger distance from the settlements, are under communal management. The fields are managed by a communal fallow and rotation system, typical for this Andean region. The crop rotation is collective in order to keep animals out of the fields under cultivation. The communal land is divided into a certain number of sectors corresponding to the number of years of the crop rotation including fallow years. The community decides collectively on the crop rotation for each sector. Each farm household holds temporary user rights for several fields in each sector. However, during fallow periods the temporary user rights are transferred back to the community, and the fields are used as pastures with communal access. These fields tend to be extensively cultivated, as the crops are exposed to climatic risks and theft (Bernet 1995). The profitability of soil conservation practices is likely to be higher on the more intensively cultivated fields in the lower parts of the catchment than on the extensively cultivated fields in the upper parts, as the agricultural production tends to have higher financial value on the former fields than the latter. However, implementing soil conservation practices at the upper part of the catchment also has various benefits, for example, improvement of soil quality over the long term, conservation of rainwater so crops suffer less stress due to drought, and reduction of surface runoff causing soil erosion further downhill.
Data Collection and Variables Used in the Empirical Model
In January and February 2002 a survey was carried out among 176 farm households, which were randomly selected. Local agricultural engineers were contracted to interview the farm households in the local language, Quechua, using a structured questionnaire written in Spanish. A quality control on the interviews and completed questionnaires was carried out by one of the authors. Cross-sectional data was collected on farm household characteristics, the farming system, farmland characteristics, agricultural production, program participation, use of soil conservation practices, and the perception and opinion of the farm household on soil conservation practices. Table 4 gives a cross table of program participation and adoption decisions.
Program Participation and Soil Conservation Adoption
This table gives some initial insight into the relation between program participation and soil conservation adoption. The majority of people who do not participate in any program also do not adopt (92.5%). However, the four nonparticipants who did adopt show that program participation is not a prerequisite for adoption. A second interesting point is that of the 22 farmers who solely participated in PRONAMACHCS, only 12 adopted soil conservation measures and 10 did not. For MARENASS the relation between program participation and adoption seems to be much stronger, with 55 of the 62 participants adopting (88.7%). Farmers that participated in both programs also widely adopted soil conservation practices (86.8%). The overall rough picture suggests that participation in MARENASS is strongly related to soil conservation adoption, whereas participation in PRONAMACHCS is not. The question remains however, whether there is a causal relation between program participation and soil conservation adoption. The trivariate probit estimates should cast light on this question.
Based on the findings of theoretical models and empirical studies (see introduction), several variables were selected to explain the decisions on program participation and adoption (Table 5). Additional variables were included in order to reveal whether soil conservation fits within the farm household income strategy and the actual need for soil conservation based on the characteristics of the household's farmland. Equation [1] indicates that the decision to participate in a soil conservation promotion program depends on the perceived gain in utility from participation. Whether a farm household perceives to gain utility from program participation is expected to depend on the characteristics of the (heads of the) households (gender, age, education, risk perception, time horizon in decision making, market orientation, presence, social network) and characteristics regarding the importance of its farming activities for the household (ratio off-farm to farm income, farm size, ratio of farmland located in the valley to total farmland). The same set of independent variables is used in the equations for participation in PRONAMACHCS and MARENASS. It is assumed that farm households with less labor available are less likely to get involved in program activities. It is also assumed that the labor availability of farm households is defined by the gender of the head of household (as female-headed households normally have less labor available), farm household size, and the presence of the head ofhousehold. Seasonal migration by men is an important means in this area to obtain additional cash income. However, if the male head of household is away for a few months a year when the agricultural season is finished, the household is less likely to get involved in program activities. As both programs are oriented toward agricultural productivity, in particular arable crops, the importance of arable farming for the farm household is also likely to influence the perceived utility of program participation by farm households. Households that are financially more dependent on farm income (that is, have a low value for the ratio off-farm to farm income) are likely to be more interested in improving agricultural production and thus program participation. Similarly, farm households that are more market oriented, that is, farm households that sell a larger proportion of their produce on the local market, are assumed to be more interested in program participation. Farm households with more land located in the valley also tend to live near the road and have a better access to markets and information sources, which might facilitate program participation. The characteristics of the heads of household define their perception of the utility gain (or loss) of program participation. Older heads of household might be less interested in getting involved in new activities, whereas heads of household with a higher level of education might be more interested (e.g., Feder, Just, and Zilberman 1982). Depending on the incentives used by programs, farm households with an aversion to risk and a shortterm time horizon might be interested in one program, but not the other. Participants of PRONAMACHCS are sure they will receive tools in return for their labor. Participants of MARENASS are less sure about their rewards, as this depends on their performance in the competitions. However, PRONAMACHCS distributes the tools only at the end of each year, whereas MARENASS organizes the farmer competitions every three months. PRONAMACHCS participants thus have to wait longer to receive the rewards than MARENASS participants. It is expected that farm households with a larger social network are more likely to be involved in new program activities because of their social network.
Description of Explanatory Variables Used in Explaining Program Participation and Soil Conservation Adoption
The explanatory variables expected to determine the decision to adopt soil conservation practices include program participation (PRONAMACHCS or MARENASS or both), farmland characteristics (total farm area; ratio of farmland without access to irrigation to total farmland; ratio of farmland with no slope to total farmland; ratio of farmland without stones to total farmland, as a proxy for soil degradation; average distance between fields and house), and household characteristics (household size; gender, age and education of head of household; risk behavior; time horizon; market orientation). As explained above, participation in the programs PRONAMACHCS or MARENASS is expected to positively influence the adoption decision as program participation influences the perception of farm households on soil erosion (e.g., Ervin and Ervin 1982). It is also expected that farm households with more farmland are more likely to adopt soil conservation practices (e.g., Feder, Just, and Zilberman 1982). Farm households with access to irrigation are more likely to adopt soil conservation practices as well, as the combination of terraces and irrigation allows a household to cultivate crops during the dry season on sloping land. The degree of soil degradation, and thus the need for soil conservation, is expected to positively influence the adoption decision (e.g., Lynne, Shonkwiler, and Rola 1988). Farm households with a larger proportion of farmland with stones are thus expected to adopt soil conservation practices, as land with stones is normally a sign of soil degradation. Farm households with a larger proportion of flat farmland, on the other hand, are less likely to adopt soil conservation practices, as these are normally implemented on sloping land. If the average distance between the fields of the farm household and its homestead is larger, adoption is less likely, since the implementation of soil conservation practices requires a huge amount of labor input. Farm household size might affect adoption, as limited labor availability can be a constraint to the implementation of soil conservation practices (e.g., Nowak 1987). Male-headed farm households are therefore expected to be more likely to adopt soil conservation practices. Head of households with higher education are also expected to be more inclined to adopt new technologies, as they might acquire new knowledge more easily. Risk-averse farm households and those with a short-term time horizon are less likely to adopt soil conservation practices, as the benefits of these practices are generally not immediately tangible and are instead achieved over the long term (e.g., Marra, Pannell, and Ghadim 2003). Farm households that sell large proportions of their agricultural produce at local markets are more likely to adopt soil conservation practices, as these practices have the potential to increase agricultural production.
The variables Education, Social network, Risk taker, and Long term are explained in further detail:
Education. The education level of the head of farm household is ranked as follows: Education 5 1 if no education; 5 2 if primary school unfinished; 5 3 if primary school finished; 5 4 if secondary school unfinished; 5 5 if secondary school finished; 5 6 if higher education. A drawback of this ranking is that it treats education as a cardinally measured variable, whereas the ranking is in fact ordinal. As a consequence, each incremental step of education is assumed to have the same marginal effect on adoption. The advantage is, of course, that different levels of education can be captured by one variable. The less desirable alternative would have been to include five dummy variables to represent these different levels of education.
Social network. The head of a farm household is considered to be more actively involved in the social life of the community if he or she performs a social duty such as being a member of the community council or a committee.
Risk taker. The head of household was asked to choose between two different deals: a risky deal with a chance of high profit or loss, and a deal with sure but moderate profit. Expected profit of the two deals was the same. In the case where the head of household chose the risky deal, the variable has value 1, and 0 otherwise.
Long term. The head of household was asked to choose between two deals: more money in the long term, or less money in short term. Again, the expected profit of the two deals was the same. The variable has value 1 if the head of household chose the long-term deal, and 0 otherwise.
VI. Empirical Results on Program Participation and Soil Conservation Adoption
Table 6 presents the estimation results of the trivariate probit model used to explain participation in the programs PRONAMACHCS and MARENASS and soil conservation adoption. The Wald test statistic of 181.6 exceeds the critical 
value of 58.12, indicating that thenull hypothesis that all slope parameters βj are jointly equal to zero is rejected.
Trivariate Probit Model for Program Participation and Soil Conservation Adoption
The last lines of Table 6 show that all correlation coefficients of the residuals of the three decisions are not significantly different from zero. This implies that the decisions to participate in PRONAMACHCS, to participate in MARENASS, and to adopt soil conservation measures are made independently by the farm households. As PRONAMACHCS and MARENASS are two different programs (topdown versus participatory, conservation versus livelihood improvement), it is plausible that these participation decisions are made independently. It also implies that there are no common effects missing in PRONAMACHCS participation or MARENASS participation and soil conservation adoption, suggesting absence of self-selection.
Participation in PRONAMACHCS
The significant explanatory factors for the decision to participate in PRONAMACHCS are gender of household (+for male heads), social network (+), farm size (+), and ratio of farmland located in the valley (+). Farm households with more farmland are more likely to participate in PRONAMACHCS. In general, farmers with more land are more oriented toward agricultural production and might therefore be more interested in this program. PRONAMACHCS participants also have more land located in the valley bottom. This seems counterintuitive, as these farm households have relatively less sloping land. However, these farm households often live in the lower parts of the catchment, near the main road. According to the farmers, the technical staff of PRONAMACHCS stay near the roads in the valley when they visit the community, and use the horn of their motorbikes to inform the people that they have arrived. Apparently not much effort is made to reach the farm households in the other parts of the catchment. Due to its compulsory weekly activities, PRONAMACHCS is mainly attractive for farm households with more labor available. Female-headed households often struggle to provide labor on a regular basis and are thus less inclined to participate. Farm households with a stronger social network are more actively involved in community life. Therefore, these households are more integrated in the community network and thus have better access to information about new events and activities, resulting in higher participation rates.
Participation in MARENASS
The significant explanatory factors for the decision to participate in MARENASS are social network (+), risk-taking attitude (-), market orientation (+), and off-farm/ farm income ratio (-). These results show that risk-averse farm households are attracted to MARENASS, suggesting that these farm households consider the activities of MARENASS as risk reducing. Market-oriented farm households are more likely to participate in MARENASS. These farm households may anticipate benefits in terms of increased agricultural production, meaning increased income, from participation in MARENASS. This is confirmed by the negative sign for income ratio, meaning that farm households that rely more on farming activities for their cash income are also attracted to MARENASS. MARENASS's strategy of transferring funds and responsibility to the communities might also be more attractive to farm households that are more market oriented. Social network is an explanatory factor for participation in MARENASS, as well as in PRONAMACHCS. This stresses the importance that programs take account of, and strengthen, the social networks within communities. However, another explanation might be that farm households with a strong social network are different in their attitude and behavior than those with a limited social network. These characteristics might also explain their interest in getting involved in programs.
Adoption Decision
A significant factor in the adoption decision is the need for soil conservation practices: farm households with more sloping farmland are more inclined to adopt soil conservation practices. Considering program participation, the results show that MARENASS has a significant causal impact on adoption of soil conservation practices, whereas PRONAMACHCS does not. In other words, the characteristics and setup of MARENASS are effective in promoting adoption of soil conservation practices at the household level, whereas those of PRONAMACHCS are not. This can be explained by the fact that PRONAMACHCS participants do not necessarily implement practices on their own farmland. The technical staff of PRONAMACHCS decides on the location of the practices, which might be on common land or on the land of another participant. Apparently, participants do not copy these practices on their own farmland. With MARENASS, however, participants are stimulated to implement soil conservation practices on their own farmland, as they can then obtain points in the farmer competitions. Nevertheless, there is a risk that farm households implement soil conservation practices only because of these farmer competitions, and abandon these practices afterward. Another explanation for the differences in impacts between the two programs is that MARENASS uses farmer-to-farmer extension methods to instruct farmers how to implement soil conservation practices, whereas the technical staff of PRONAMACHCS normally directs and supervises the implementation of soil conservation practices.
VII. Discussion
Variables commonly found to be significant in adoption studies, such as age, education, or risk attitude (see Table 3) were not significant in this case study. However, risk attitude did influence the decision to participate in MARENASS, and since MARENASS participation has a significant impact on adoption, risk attitude affects adoption of soil conservation practices indirectly. It was somewhat unexpected that the variables for long-term time horizon and risk attitude were not (directly) significant in the adoption equation, since the implementation of soil conservation practices is typically seen as a long-term and risky investment. It may be that the incentives provided by the programs offset the limitations of long-term and uncertain benefits sufficiently. Social network appeared to be a significant factor in the program participation decisions. Others have also found that social networks or social capital are important factors that positively influence technology adoption (e.g., Cramb 2005; Monge, Hartwich, and Halgin 2008).
Using a binary model such as the trivariate probit used here limits the analysis, as only the adoption decision is analyzed, whereas the extent of adoption is not considered. Unfortunately it was not possible to distinguish the extent of adoption attributable to the different programs for those households that participated in both programs. Therefore, it was not possible to apply a Tobit-type model to analyze adoption decisions and extent of adoption in one coherent framework, allowing for different variables in both processes and different marginal effects. This may explain why some of the variables (e.g., education, age) that were commonly found to be significant in other studies using a Tobit-type model were not significant in the model presented in this paper.
Another factor commonly found to be significant in adoption studies, the (perceived) degree of soil degradation, was not significant in our model. The variable representing the ratio of land without stones to total farmland, as a proxy for soil degradation, was not significant in the adoption equation. However, the ratio of flat land to total farmland was significant and negative, and it can therefore be assumed that farm households who have relatively more sloping fields (and are thus more vulnerable to soil erosion) are more likely to adopt soil conservation measures. However, it has to be noted that this variable is an aggregated variable of the fields at the household level, and field-level information can thus be lost. Farm households were asked to give characteristics for their individual fields. Considering a sample of all individual fields (n = 633), it becomes apparent that there is a significant positive correlation (p = 0.000, Mann-Whitney test) between fields with stones and adoption of soil conservation practices. Forty percent of the fields with stones had soil conservation practices, compared to 24% of fields without stones. Fields with stones had also significantly steeper slopes than fields without stones (p = 0.000).
Table 7 provides additional detail about the adoption effort at the household level. Mean values are given for the characteristics of soil conservation adoption at the household level for the different groups of program participants (i.e., not participating in any program, participating in PRONAMACHCS only, participating in MARENASS only, participating in both programs). Statistical tests revealed whether differences in adoption effort between the groups were significant. The differences between the PRONAMACHCS-only and the MARENASS-only groups were also compared separately in order to reveal any significant differences between the two programs. On average, households participating in both programs invested most labor in soil conservation at the household level, and households not participating in any program the least. However, the difference in labor invested in soil conservation between the households participating in either PRONAMACHCS or MARENASS only was not significant at the household level. MARENASS participants invested significantly more in soil conservation in terms of other costs in addition to labor than PRONAMACHCS participants. MARENASS participants implemented the soil conservation practices in groups (based on reciprocal assistance), and the owner of the fields typically provided refreshments for the group members, adding to the costs of construction. MARENASS participants might also have been obliged to purchase tools for terrace construction, whereas PRONAMACHCS participants received these tools for free. Households participating in both programs received most incentives, and households not participating in any program received none. The monetary value of the incentives (tools for PRONAMACHCS participants and cash for MARENASS participants) received by households in either program did not differ significantly between the two programs. There was no significant difference in the total area with soil conservation at farm household level between the different groups of program participant.
Adoption Effort at Household Level
Though MARENASS has a causal effect on the adoption of soil conservation practices at the household level, in contrast to PRONAMACHCS, this does not guarantee a more effective prevention of soil erosion at the catchment level. Table 8 gives more details on adoption effort and field characteristics at the field level. Households participating in MARENASS invest only slightly, but significantly, more labor in soil conservation at the field level than their peers participating in PRONAMACHCS only. However, PRONAMACHCS-only participants had installed a significantly larger proportion of soil conservation practices on steep sloping (hence, vulnerable to soil erosion) fields than the MARENASS-only participants. MARENASS-only participants installed more soil conservation practices in the bottom of the valleys than the PRONAMACHCS-only participants. These results suggest that farm households participating in either PRONAMACHCS or MARENASS implemented soil conservation practices at different locations in the catchment and also reveal different levels of effort.
Adoption Effort At Field Level
The different approaches of the two programs toward the implementation of soil conservation practices may result in a great difference in impact on soil erosion. The concept of disproportionality applies to situations where the average impact of one group is significantly greater than the average impact of other groups. Nowak, Bowen, and Cabot (2006) used this concept to address interactive or multiplicative effects between behavioral and biophysical differences. For example, an inappropriate, erosion-inducing land management practice may have no negative impact on the environment in a well-buffered biophysical setting, whereas a so-called best practice may be detrimental in a vulnerable biophysical setting. Comparing the characteristics of fields with soil conservation practices implemented with the assistance of PRONAMACHCS or MARENASS, it appeared that MARENASS participants put more effort into soil conservation practices but installed them on relatively flat land more often than PRONAMACHCS participants (see Table 8). This is due to the different approaches of the two programs, as the farm households participating in the different programs have a relatively similar amount of flat land.2 Though this research did not specifically address this concept of disproportionality, our results indicate that MARENASS, despite the increased effort, is likely to be less effective in reducing soil erosion at catchment scale than PRONAMACHCS. It thus appeared that MARENASS participants preferred to implement soil conservation practices on the more productive fields to boost agricultural production, rather than on the marginal fields to reduce soil erosion. From the farm household’s point of view, it makes more sense indeed to invest in fields where the effect on productivity or profitability is higher (Posthumus and Stroosnijder 2010). However, from a soil conservation perspective, one would implement soil conservation practices at a catchment scale, in particular targeting hot spots of erosion on the steeper slopes.
There seems to be an unwritten assumption that soil conservation programs can be effective in reducing soil erosion only when there is a high uptake of soil conservation practices at the farm household level (e.g., Antle and Diagana 2003; Bunch 1999; Grepperud 1995; Winters, Crissman, and Espinosa 2004). However, this study suggests that a good uptake of soil conservation practices by farm households does not guarantee an effective soil conservation program in terms of reduced erosion at the catchment scale, because of the problem of disproportionality. In order to secure their livelihoods, farm households tend to prioritize agricultural production over soil erosion, and therefore they are likely to implement soil conservation practices at different locations than recommended by a soil conservation engineer (Winters, Criss-man, and Espinosa 2004). Expert knowledge and permanent support are essential if soil erosion is to be reduced at the catchment scale. A participatory program such as MARENASS seems to be more effective in introducing technologies at the household level, but if an environmental problem at the community or catchment scale needs to be addressed, a top-down approach such as PRONAMACHCS might be more appropriate if it allows for targeting hot spots with high risks of environmental degradation.
Comparing the two programs in this case study did not result in an obvious “winner” even though MARENASS appeared to obtain better results in terms of soil conservation adoption. It depends on the objective of the program (mitigate soil degradation vs. improve farm household productivity) and the local context as to which type of program is most appropriate to obtain sustainable results. Policy makers or NGOs that want to promote soil conservation should be clear about their underlying aims. If the main aim is to conserve natural resources (benefitting society rather than individuals), large-scale and long-term interventions, such as PRONAMACHCS, might be necessary, in particular when the land is under communal management. In such a case, local authorities could play a role in the promotion of soil conservation and pay peasant communities for the environmental services they provide, especially if there are downstream benefits (Posthumus 2007; Wunder 2007). However, these agreements, well-established in Europe in the form of agrienvironmental schemes, are still rare in developing countries. On the other hand, if the main aim is to improve rural livelihoods by enhancing agricultural productivity through soil conservation, a participatory approach such as used by MARENASS might achieve better results, as farm households are encouraged to take ownership and adapt technologies to their personal circumstances. Unfortunately, the trivariate probit model could not take the performance of individual program staff into account, but discussions with farm households and program staff revealed that the attitude and skills of field staff are also important factors influencing the uptake of soil conservation practices.
VII. Conclusions
Though the importance of programs promoting soil conservation for the adoption process is widely acknowledged in the literature, this is rarely taken into account in the analysis of adoption decisions. Not considering the decision on program participation may lead to biased results in the analysis of adoption behavior. For this reason, a trivariate probit was applied to estimate the impact of participation in two different programs for the adoption of soil conservation practices by farm households in Peru. However, it has to be noted that modeling of adoption decisions is preferably cast in a limited dependent variable framework (e.g., a Tobit or sample selection model) to explain the extent of adoption versus nonadoption. Unfortunately, such a model could not be applied in this study. This complicates the comparison of the results presented in this paper with the findings of other studies that applied Tobit or sample selection models to explain adoption. Further research, including in different socioeconomic and geographical settings, is needed to confirm, or disclaim, the findings presented in this paper.
In earlier adoption studies it has been found that farm household characteristics such as education, age, labor availability, or risk-taking attitude significantly influence the adoption of soil conservation practices. In our study farm household characteristics influenced the decisions to participate in an extension program, but not the adoption decision itself. Results also show that the two programs attract different types of farm households.
The results of the trivariate probit show that the causal effect of extension programs on the adoption of soil conservation practices at the household level varies by program. PRONAMACHCS has no causal effect on the adoption decision, whereas MARENASS has. This is best explained by the different approaches used by the programs to promote soil conservation. Farmer-to-farmer extension and incentives such as farmer competitions have a positive influence on the adoption decision at the farm household level. However, this does not necessarily guarantee effective implementation of soil conservation practices at catchment scale if there is no strategy in place to target areas with high risks of soil erosion.
PRONAMACHCS participants, who were guided by technical staff, appeared to have put more effort in erosion-prone fields. MARENASS participants, who had more autonomy in the implementation of soil conservation practices, opted to invest in the more productive fields to boost agricultural production. Despite the raised awareness on soil erosion, farm households rated productivity higher than mitigation of soil degradation in their decision to implement soil conservation practices. Although MARENASS participants appeared to put more effort into soil conservation than PRONAMACHCS participants, the effect on soil erosion at catchment scale is likely to be less because the practices were installed on fields that were less prone to erosion. It appears that MARENASS participants implemented soil conservation practices to increase agricultural production rather than to reduce soil erosion. This suggests that the objective of soil conservation (conservation of natural resources versus improvement of agricultural productivity) influences how and where these practices are implemented. Farmers' objective of agricultural productivity often exceeds soil conservation, and participatory soil conservation programs may therefore result in different outcomes than top-down programs. While local circumstances (e.g., biophysical conditions, institutional framework, or social capital) will dictate which approach is most appropriate, the results of this case study indicate that a combination of approaches (top-down interventions with continuous support versus participatory bottom-up approaches) might be required to achieve different aims (conservation of natural resources vs. improvement of agricultural productivity) through soil conservation.
Appendix
The likelihood function of the trivariate probit model is defined by
where probabilities are defined by a trivariate normal distribution function φ3(.):
Footnotes
The authors are, respectively, researcher, Natural Resources Institute, University of Greenwich at Medway, Chatham Maritime, Chatham, U.K.; assistant professor, Agricultural Economics and Rural Policy, Wageningen University, Wageningen, The Netherlands; and professor, Centre for International Development Issues, Radboud University, Nijmegen, The Netherlands.
↵1 Greene (1998) indicates that this result was already presented by Maddala (1983, 123). For an application of a bivariate probit model with endogenous binary regressors see Evans and Schwab (1995).
↵2 The outcome of the analysis of variance shows that there is no difference in mean values of ratios of flat land to total farmland between the participant groups. The F-value was 0.33, which is lower than the critical value of 2.53, meaning that the null hypothesis (the mean values are equal) is accepted.
References
In this issue
Jump to section
- Article
- Abstract
- I. Introduction
- II. Soil Conservation Programs in Peru
- III. Conceptual Framework for Soil Conservation Adoption
- IV. A Model to Explain Program Participation and Soil Conservation Adoption
- V. Methodology
- VI. Empirical Results on Program Participation and Soil Conservation Adoption
- VII. Discussion
- VII. Conclusions
- Appendix
- Footnotes
- References
- Figures & Data
- Info & Metrics
- References